Research

My research aims to develop machine learning systems that are adaptable, robust and energy‑efficient. I investigate continual and domain‑incremental learning methods, robust and secure ML techniques, federated and distributed learning algorithms, multimodal perception systems and explainable AI (XAI) approaches. By combining systems and algorithmic perspectives, I build models that generalize across different devices, domains and over time, while maintaining high performance under constraints like limited data, hardware resources and adversarial conditions. A recent focus of my work is on XAI techniques that provide transparency and interpretability in model decisions.

Some of my current interests include:

• Continual learning frameworks to mitigate catastrophic forgetting.
• Robust and secure ML systems resilient to data poisoning and adversarial attacks.
• Federated and distributed learning methods for resource‑constrained devices.
• Energy‑efficient and multimodal perception systems for edge and robotics applications.
• Explainable AI (XAI) methods to interpret and visualize model decisions.

Publications

• Overcoming Catastrophic Forgetting with Continual Learning for Smartphone‑based Indoor Localization – Introduced the first path‑based class‑incremental learning framework for localization. IEEE Access, 2025.
• Domain‑Incremental Learning for Indoor Localization using Smartphones – Proposed a novel domain‑incremental framework addressing temporal and device‑induced shifts. IPIN Conference, 2025.
• Overcoming Data Poisoning Attacks in Heterogeneous Federated Machine Learning for Indoor Localization – Developed a gradient‑saliency based adversarial detection mechanism. DATE, 2025.
• Unified Class and Domain Incremental Learning with Mixture‑of‑Experts for Indoor Localization – Developed a domain‑incremental VAE‑driven representation learning framework. DATE, 2026.

Experience

• Applied AI/ML Research Solutions Intern – Nokia Bell Labs (Jun 2025 – Aug 2025)
  Developed continual object detection systems extending YOLOv7/YOLOv8 with regularization and knowledge distillation; optimized for edge deployment with low storage and fast inference; contributed to multimodal sensing for indoor robotics.
• Graduate Research Assistant – CSU Vice President of Research (Mar 2024 – May 2025, Sep 2025 – Present)
  Supported ML researchers in distributed training, Slurm optimization and GPU profiling; built automated workflows and reproducible environments; developed tools for parallel training, profiling and dataset preprocessing on HPC hardware.
• Senior Systems Engineer – Infosys Ltd. (Jan 2021 – Dec 2022)
  Engineered CI/CD pipelines for large‑scale C++ systems; developed Go benchmark suites for server performance profiling; created a Node.js‑based telemetry and monitoring system for production workloads.

Projects

• Adversarial Data Poisoning Attacks in NL‑to‑Code Generation – Investigated targeted and untargeted poisoning attacks on NL‑to‑code LLMs (CodeT5+, CodeBERT, StarCoder) to reveal robustness gaps and defensive strategies.
• Federated Learning for Indoor Localization – Achieved 7.8× accuracy improvement and 2.1× latency reduction using fused models and saliency-driven aggregation.
• Continual Learning for Indoor Localization – Enhanced accuracy by 58.6%, reduced storage by 74.9%, and decreased energy use by 69.53% via continual learning strategies.
• Dog Emotion Detection (Edge ML System) – Deployed a CNN on Raspberry Pi using TFLite for real‑time emotion classification with 83% accuracy.

Contact

Email: akhil.singampalli@colostate.edu

Location: Fort Collins, CO, United States

If you would like to collaborate or learn more about my research, feel free to reach out via email.